fix(recording): 录音条出现时 mic 已 capture，不再吞开头字

[appergb]

User description

用户反馈

按 Option 键后已经看到录音条，但开口的前几个字被吞掉，应该立即开始录音。如果当时 ASR 没连上不要紧，按顺序发送即可（这部分 `DeferredAsrBridge` 已经做了，本次不动）。

根因

`coordinator.rs::begin_session()` 在 `line 1128` 就 `emit_capsule(Recording)`，但 `Recorder::start` 还要再走 ~57 行（read creds / new ASR / new bridge / acquire mute / cpal init），其中 cpal init 在 macOS 上有 50-200ms 启动延迟。

窗口期内：

• 视觉：录音条已显示 → 用户以为可以说话了
• 物理：mic 还没产生第一帧 PCM
• 用户开口的前几个字 物理上无法被 capture（不是丢字，是没录到）

`DeferredAsrBridge` 已经覆盖 「recorder 已开 → ASR 没连上」的间隙（缓存 PCM 等 ASR attach 后按顺序后送），但 「录音条已显 → recorder 没开」之前没人覆盖。

修复（surgical）

`emit_capsule(Recording)` 从 `begin_session()` 挪到 `start_recorder_for_starting()` 内部 `Recorder::start` 成功的 `Ok` 分支：

```diff

• emit_capsule(inner, CapsuleState::Recording, 0.0, 0, None, None); // 提前 emit 删除

• // 不在这里 emit Recording：让 start_recorder_for_starting 在 Recorder::start

• // 成功后再发，确保「录音条出现」时 mic 已经在 capture。

  match Recorder::start(consumer, level_handler) {
  Ok((rec, runtime_errors)) => {
  ...
  log::info!("[coord] recorder started ...");

•        // ★ 录音器实际启动后再发 Recording capsule
  

•        emit_capsule(inner, CapsuleState::Recording, 0.0, 0, None, None);
     }
  

```

三条 ASR 路径（Local Qwen3 / Whisper / Volcengine）共用 `start_recorder_for_starting`，自动受益。代价是用户按热键到看到录音条多 ~50-200ms 延迟，但「录音条出现 = mic 100% 在 capture」体感诚实。

不动的部分（已合规）

• `DeferredAsrBridge` 按顺序缓冲 PCM + ASR attach 后 drain — 保留
• `finish_starting_session` cancel race 检查 — 保留
• error / cancel 路径的 emit_capsule(Error) — 保留

Test plan

• `cargo check` 通过
• CI `Windows Tauri checks` + `pr_agent_job`
• 真机 mac：按 Option 立刻说"测试一二三"，转录出"测试一二三"完整无丢
• 真机 mac：按 Option 立刻看到录音条 + 50-200ms 延迟体感可接受

---

PR Type

Bug fix

---

Description

• 延后 Recording capsule 发送时机

• 改为录音器启动成功后再显示录音条

• 增加 stop/cancel 竞态保护

• 避免短按热键时状态闪回

---

Diagram Walkthrough

flowchart LR
  A["begin_session"] -- "prepare session" --> B["start_recorder_for_starting"]
  B -- "Recorder::start succeeds" --> C["emit Recording capsule"]
  B -- "stop/cancel pending" --> D["skip overwrite"]
  C -- "mic already capturing" --> E["user speaks safely"]

Loading

File Walkthrough

	Relevant files
Bug fix	coordinator.rs Delay recording capsule until mic starts                                  openless-all/app/src-tauri/src/coordinator.rs Removed the early emit_capsule(Recording) from begin_session(). Emit Recording only after Recorder::start returns Ok. Added a state check for Starting, pending_stop, and cancelled. Prevents overwriting stop/cancel transitions during startup. +24/-2

---

[github-actions]

PR Reviewer Guide 🔍

(Review updated until commit 5c5b575)

Here are some key observations to aid the review process:

⏱️ Estimated effort to review: 2 🔵🔵⚪⚪⚪

🧪 No relevant tests

🔒 No security concerns identified

⚡ No major issues detected

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 02dd8c56ea

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Avoid re-showing Recording after a queued stop

When the user stops or cancels while Recorder::start is still initializing, the stop path can already have emitted Transcribing/Cancelled or moved the session out of Starting; this new unconditional emit then overwrites the UI back to Recording after stop_recorder_if_pending_start_stop runs, and on the Volcengine path it can remain that way while open_session().await continues. Gate this emit on the session still being the active Starting session with no pending stop/cancel, or emit it before the pending-stop handling only when the recorder is actually going to keep running.

Useful? React with 👍 / 👎.

[appergb]

@codex P2 已采纳：commit 5c5b575

修法按你的建议：

1. emit_capsule(Recording) 挪到 stop_recorder_if_pending_start_stop 之前
2. emit 之前在同一 lock 内检查 session_id == session_id && phase == Starting && !pending_stop && !cancelled；只有 recorder 真的会继续运行时才 emit
3. 不满足条件就让 stop / cancel 路径接管 UI 信号，不再无条件覆盖

短按场景下不再出现 Recording 闪烁覆盖 Transcribing/Cancelled 的问题。

[chatgpt-codex-connector]

To use Codex here, create an environment for this repo.

[github-actions]

Persistent review updated to latest commit 5c5b575